A recent study published in The BMJ employed a BERT-based 'scientific spam filter,' which was trained on 2,202 retracted papers from paper mills, to evaluate 2.6 million cancer studies conducted between 1999 and 2024. The research, led by biostatistician Adrian Barnett from Queensland University of Technology, flagged 261,245 papers, representing 9.87%, as exhibiting suspicious writing patterns, a significant increase from approximately 1% in earlier years.